基于多属性深度学习的相控约束建模反演技术OA

In areas with low well control,low-frequency models constructed via well-log interpolation are prone to the"bull's-eye"artifact.This leads to significant deviations between the model and actual subsurface geological conditions,ultimately reducing the accuracy of seismic inversion results.Taking a block in the Songliao Basin as a case study,this paper first calculates the sand-to-strata ratio(SSR)for individual wells within the target interval.Subsequently,a correlation analysis is performed between various seismic attributes and well-based parameters to select attributes with high correlation.A quantitative mapping relationship is then established between these optimal seismic attributes and the well-based SSR.A deep learning algorithm is employed for fitting calculations,ensuring that the coincidence rate between the predicted SSR and measured well data exceeds 85%.This trained mapping relationship is then extrapolated across the entire study area to generate a planar distribution map of the SSR for the target interval.By integrating the sedimentary facies interpretations from individual wells with the planar SSR characteristics,a sedimentary facies map of the target interval is compiled.This map serves as a constraint for facies-controlled post-stack impedance inversion.The inversion results show an overall coincidencerate of 86%with the sand bodies encountered in the wells,effectively validating the feasibility and application of this method for the fine prediction of sand bodies in areas with sparse well control.